Abstract:
Every day millions of users use trip planners (or equivalently journey planners) for smooth and convenient daily commuting using public transport. These trip planners (e.g., Google Maps trip planner) allow users to search for an optimal means (e.g., fastest or minimum switches) of traveling between two locations in the city using public transport. In these travel plans, a single trip may use a sequence of several modes of transport based on the available transport networks and the timetables of the services. On top of that, an Intelligent Transport System (ITS) provides data-driven services to maximize the efficiency of vehicles and the overall convenience of travelers. These services often include automated trip planning and estimating the travel time of the trips. In this thesis, we take a step towards an intelligent trip planner, that finds the most popular trip from historical trips and calculates the distribution of travel time of the trip. Given a source, a destination and a departure time, our proposed system can integrate user-defined constraints such as time, minimum switches, or preferred modes of transport. To solve the most popular trip and its variants, we propose a multi-stage deep learning architecture PathOracle that consists of two major components: KSNet to generate key stops, and MPTNet to generate popular path trips from a source to a destination passing through the key stops. To tackle the travel time estimation problem in public transport, we separately predict the distributions of times taken by a vehicle and times waiting for a vehicle. To estimate probability distributions, we introduce two approaches: PDistNet for explicit parameter estimation and SDistNet for implicit estimation. We also introduce a unique representation of stops using Stop2Vec that considers both the neighborhood and trip popularity between stops to facilitate accurate path planning. We present an extensive experimental study with a large real-world public transport-based commuting Myki dataset of Melbourne city, and demonstrate the effectiveness of our proposed approaches.
